Author Contributed Representation for Scholarly Network

Scholarly network analysis is a fundamental topic in academia domain, which is beneficial for estimating the contribution of researchers and the quality of academic outputs. Recently, a popular fashion takes advantage of network embedding techniques, which aims to learn the scholarly information into vectorial representations for the task. Though great progress has been made, existing studies only consider the text information of papers for scholarly network representation, while ignoring the effects of many intrinsic and informative features, especially the different influences and contribution of authors and cooperations. In order to alleviate this problem, in this paper, we propose a novel Author Contributed Representation for Scholarly Network (ACR-SN) framework to learn the unique representation for scholarly networks, which characterizes the different authors’ contribution. Specifically, we first adopt a graph convolutional network (GCN) to capture the structure information in the citation network. Then, we calculate the correlations between authors and each paper, and aggregate each embedding of authors according to their contribution by using the attention mechanism. Extensive experiments on two real world datasets demonstrate the effectiveness of ACR-SN and reveal that authors’ contribution to the paper varies with the corresponding authorities and interested fields.